LastPass Employee Foils AI Voice Deepfake of CEO Karim Toubba (2024)

What Happened

In April 2024, LastPass disclosed that one of its employees had been targeted in a social engineering attempt that used an audio deepfake to impersonate the company's CEO, Karim Toubba. The employee received a series of WhatsApp communications, including missed calls, texts, and at least one voicemail, from an account posing as Toubba. The voicemail featured AI-generated audio designed to sound like the CEO, applying pressure through a sense of urgency.

The employee did not comply. Because the contact arrived over WhatsApp, which is not a normal LastPass business communication channel, and because it carried classic social engineering hallmarks such as forced urgency and contact outside normal business hours, the employee grew suspicious, ignored the messages, and reported the incident to the internal security team.

LastPass stated plainly that there was no impact to the company. The security team was able to mitigate and, notably, LastPass chose to publish the incident on its own blog (authored by senior principal intelligence analyst Mike Kosak on April 10, 2024) specifically to raise awareness that deepfake tools are now cheap and accessible enough to be used in run-of-the-mill executive-impersonation fraud, not just by nation-state actors. Press reporting (BleepingComputer, SecurityWeek, PCMag) noted the cloned voice was likely generated from publicly available recordings of Toubba, such as podcast or video appearances.

This was an attempted attack that was blocked at the human layer. It also stands out because LastPass, a security company that had suffered major breaches in 2022, transparently used its own near-miss as a teaching case for the wider industry.

How the Attack Worked

Recon: the attacker obtained the target employee's contact details and identified LastPass's CEO as a credible impersonation target, drawing on the CEO's public profile and publicly available audio of his voice. Contact: the approach came through WhatsApp, a channel outside sanctioned corporate communications, using an account set up to appear as the CEO. Rapport and exploitation: the attacker layered multiple touches (missed calls, texts, and a voicemail) and used an AI-generated clone of the CEO's voice to lend authenticity, paired with manufactured urgency to push the employee to act quickly and without verification. Payout: the intended end state was executive-impersonation fraud, but the chain broke at the exploitation stage. The employee recognized the combination of an unusual channel, off-hours timing, and pressure tactics as a social engineering attempt, disengaged, and escalated to internal security, which neutralized the threat. Described at awareness altitude only, this record does not include any voice-cloning or setup instructions.

The Lure & the Tell

Pretext: an urgent request from someone presenting as the company CEO, delivered via a cloned voice to make the demand feel personal and legitimate. Visible red flags (tells): contact through WhatsApp rather than approved internal channels, outreach outside normal business hours, forced urgency pressuring an immediate response, and an unsolicited, out-of-band request from a senior executive to a mid-level employee. Any one of these should trigger out-of-band verification; together they were enough for the employee to reject and report the contact.

Outcome

Attempt fully blocked. No funds lost, no systems accessed, no impact to LastPass. The employee ignored the messages and reported to internal security; LastPass mitigated the threat, shared indicators with intelligence-sharing partners and peer security companies, and published the incident publicly to raise industry awareness.

Why It Matters

This is a clean example of an AI voice deepfake attack defeated at the human layer, showing that security awareness and a strong reporting culture remain effective even against synthetic-media attacks. It also marks a shift the industry had been warning about: deepfake voice fraud is no longer confined to nation-state actors or headline mega-losses (like the 2024 Hong Kong $25M video-deepfake case), but is now a commodity tactic used in everyday executive-impersonation attempts. That a security vendor itself was targeted, and chose to disclose a mere attempt with no losses, makes it a rare, citable teaching case.

Defenses

Out-of-band verification: confirm any urgent or unusual executive request through an established, approved internal channel before acting, never through the channel the request arrived on. Channel discipline: treat requests arriving via non-sanctioned channels (personal WhatsApp, SMS) as inherently suspect. Recognize social engineering hallmarks: forced urgency, off-hours contact, and pressure to bypass process. No-blame reporting culture: make it easy and expected for employees to escalate suspicious contacts to security. Awareness training that specifically covers AI voice cloning, so a familiar-sounding voice is no longer treated as proof of identity. Executive-impersonation playbooks and code words / verification protocols for high-stakes requests.